MIT's LOBSTgER Project: Harnessing AI and Underwater Photography to Illuminate the Gulf of Maine's Marine Life

Introduction

The Gulf of Maine, a biologically diverse marine ecosystem, is experiencing rapid environmental changes due to accelerated warming. (seagrant.mit.edu) To document and visualize these changes, MIT Sea Grant has launched the LOBSTgER project—Learning Oceanic Bioecological Systems Through Generative Representations. This initiative merges generative AI with underwater photography to create realistic visualizations of marine life, enhancing scientific storytelling and public engagement. (seagrant.mit.edu)

The LOBSTgER Project: Bridging Art, Science, and Technology

LOBSTgER is co-led by underwater photographer Keith Ellenbogen and MIT mechanical engineering PhD student Andreas Mentzelopoulos. The project trains generative models exclusively on a curated library of Ellenbogen’s underwater photographs, ensuring the resulting imagery maintains both visual integrity and ecological relevance. (seagrant.mit.edu)

Overcoming Challenges in Underwater Photography

Underwater photography in New England’s coastal waters is notoriously difficult due to limited visibility, swirling sediment, bubbles, and unpredictable marine life movement. (seagrant.mit.edu) Ellenbogen’s comprehensive dataset, captured through the “Space to Sea: Visualizing New England’s Ocean Wilderness” project, provides the foundation for training LOBSTgER’s generative AI models. (seagrant.mit.edu)

Enhancing Visual Narratives with AI

LOBSTgER’s custom diffusion models replicate not only the biodiversity documented by Ellenbogen but also his artistic style. By learning from thousands of real underwater images, the models internalize fine-grained details such as natural lighting gradients, species-specific coloration, and atmospheric textures. (seagrant.mit.edu) The models can generate new, synthetic, scientifically accurate images unconditionally and enhance real photographs conditionally, allowing for the recovery of detail in turbid water and adjustment of lighting to emphasize key subjects. (seagrant.mit.edu)

Implications for AI Image and Video Generation

The LOBSTgER project exemplifies the potential of AI in enhancing visual narratives, particularly in challenging environments like underwater ecosystems. By integrating AI into the photographic workflow, the project not only documents marine life but also reimagines how environmental stories can be told. (seagrant.mit.edu)

Exploring AI-Generated Imagery with PixelDojo

For those interested in exploring AI-generated imagery, PixelDojo offers a suite of tools that align with the techniques used in the LOBSTgER project:

• Stable Diffusion Tool: Allows users to generate high-quality images from textual descriptions, similar to how LOBSTgER creates realistic visualizations of marine life.

• Image-to-Image Transformation: Enables users to enhance existing photographs by adjusting elements like lighting and detail, akin to LOBSTgER’s conditional enhancement of underwater images.

• Text-to-Video Tool: Facilitates the creation of dynamic video content from textual input, offering a way to animate AI-generated imagery and bring static images to life.

By utilizing these tools, users can experiment with AI-driven image and video generation, gaining insights into the processes that power projects like LOBSTgER.

Conclusion

MIT Sea Grant’s LOBSTgER project stands at the intersection of art, science, and technology, demonstrating how generative AI can expand scientific storytelling by building on field-based photographic data. As AI continues to evolve, its integration with traditional documentation methods offers new avenues for visualizing and understanding our changing world.